基于LSTM-MLP的输电线路触树故障诊断研究OA

[Objective]Transmission lines often traverse forest areas,where tree growth narrows the tree-line distance to the safety threshold,and faults are easily triggered when superimposed with short-term environmental disturbances such as wind deflection and sag variation,which seriously threatens power supply safety and public security.Aiming at the problems of untimely early warning and high manual inspection cost of transmission line tree-contact faults,this paper proposes a tree-contact fault diagnosis method based on LSTM-MLP.[Methods]Firstly,through the derivation of the π-type equivalent model of transmission lines,this study reveals the functional relationship between equivalent susceptance and tree-line distance,as well as the nonlinear correlation between susceptance and line operation data(voltage,current,power,etc.),which provides a physical basis for fault feature extraction.Secondly,to solve the problem that susceptance cannot be directly measured,a simplified estimation formula of equivalent susceptance is derived,and an LSTM-MLP hybrid model is constructed.The LSTM is used to capture temporal evolution characteristics,and the MLP is used to realize nonlinear mapping,forming a fused input mode of"11-dimensional operation data+susceptance feature".Finally,the proposed method is validated based on the annual operation data of a 220 kV transmission line in forest areas of Northeast China and 3 valid tree-contact fault cases.[Results]Experimental results show that under the current data conditions,the model can accurately identify the fault risk 24 hours before the fault occurs.After integrating the susceptance feature,the diagnostic accuracy reaches 92.3%,which is significantly better than comparative models including RF and LSTM.[Conclusion]This method breaks through the limitations of traditional research on transmission line tree-contact faults,addresses the core drawbacks of the conventional fixed threshold method,namely poor environmental adaptability and insufficient early warning lead time,effectively reduces the false alarm and missed alarm rates of faults,and achieves accurate advance prevention and control of tree-contact faults.The collaborative framework of"physical mechanism constraint+data-driven learning"constructed by this method takes into account both the physical constraints of the data-driven model and the working condition adaptability of the mechanism-based model,providing solid technical support for the early warning of such faults.Furthermore,this method can be directly deployed based on the existing Supervisory Control And Data Acquisition(SCADA)data of substations without additional equipment,which effectively reduces operation and maintenance(O&M)costs and field operation risks.It is capable of effectively preventing large-scale blackouts and forest fires caused by transmission line tree-contact faults,reliably ensuring the safe and stable operation of the power grid,and thus has extremely high practical engineering value and industrial application prospects.